People vs. Machina sapiens. Trial Part 1: The case for the prosecution.

Dateline: June 22, 1997

PHILOSOPHY (and/or religion) is what’s left when science fails to provide answers to our questions about how and why things are the way they are and work the way they work. Modern science is able to answer questions at a fast and accelerating clip, with the result that philosophy is getting squeezed into a tighter and tighter corner. Science has better answers than philosophy for why the sky is blue, whether the earth is flat, how the Universe works, and why it tends not to rain if you take an umbrella. The last battles are about the human mind and the mind of God. And science seems bound to win.

"There’s no reason why a computer that’s simulating the way the neurons in the brain work won’t be intelligent," reluctantly conceded philosopher Hubert Dreyfus on American TV in May 1997, after three decades of dismissing the claims of what he dubbed "strong AI." Professor Dreyfus should have stuck to his guns. He was right the first time. A "computer simulating the way neurons in the brain work" is not intelligent. The scientific evidence is mounting that only a whole organism—an embodied mind—can be intelligent, and that there’s a lot more to a body than a plastic case with holes for keyboard input and monitor output.

John Searle is another famous AI hype-watcher. He is most well known for his "thought experiment" known as the Chinese Room Experiment, which ostensibly disproved strong AI’s claim that an algorithm—a step-by-step procedure for solving a problem—could be or become intelligent in and of itself. I’m not going to describe the experiment—plenty of others have done so, and it is enough to note that there exist several well-crafted objections to its premises (see the link just provided). What I wish to point out is that, like Dreyfus, Searle was essentially right at a time when symbol-manipulating algorithms seemed to be all there was to computing. But times have changed.

Oxford mathematician Sir Roger Penrose finds Searle’s Chinese Room argument attractive but not definitive. It "has a considerable force to it, even if it is not altogether conclusive," he wrote in The Emperor’s New Mind. His main problem is with the strong AI claim that it is possible for an algorithm to "experience feelings; have a consciousness; be a mind"—all by itself. Most AI folks today would share Penrose’s skepticism of this claim, I think; again on the basis that mind must be embodied to exist. Penrose also disputes the claim, shared (he says) by John Searle and the strong AI guard (strange bedfellows!), that everything—life, the universe, the girl next door, etc.—is a digital computer. While we can certainly demonstrate that a brain can do some computer-like things and computers can do some brain-like things, we just don’t know, says Penrose, enough about the physics of consciousness to claim that mind and computer are fundamentally identical.

What The Emperor’s New Mind is leading up to, after nearly 400 pages of dubiously necessary advanced mathematics (Douglas Hofstadter’s Gödel, Escher, Bach does a much better job, in my opinion, of both explaining the math and relating it to mind), is that our conscious (waking, aware) minds might result from non-algorithmic quantum processes, while the subconscious operates like a computer program—algorithmically. (See last week's feature for an overview of quantum processing.)

It’s astonishing what the algorithm known as genetic code in the human body can achieve. For example, we now know that women’s predisposition to sociability and intuition (and the relative lack thereof in men) is programmed into us at birth. These traits are therefore inherent not primarily to individuals but to the species, and this suggests that species-level traits (which are nevertheless observable in an individual's behavior) are subconscious and algorithmic, thus supporting at least part of Penrose’s argument. Our ability at the conscious level to suppress these traits may or may not support the other part of Penrose’s argument for a non-algorithmic conscious mind.

Though Penrose denies setting up a straw man, in effect this is what he achieves in focusing so heavily on the strong AI notion of intelligence based on algorithms isolated from environment. I found no mention in his book on the importance of heuristics, for example, in decision making. (I skipped rapidly past the math which forms the bulk of his book, so it’s possible I missed it, but heuristics is a term not listed in the index.) He seems to me to capture some of the essence of heuristics, though, in his belief that "aesthetic" criteria govern subconscious and conscious judgements and decisions.

He also fails to consider the possibly key role of emergence in the evolution of intelligence, consciousness, and mind. He agrees it to be conceivable that we could create an intelligent machine "if we ever do discover in detail what quality it is that allows a physical object to become conscious," and is awed by the power of natural selection, but appears to miss the importance of emergence.

Dreyfus, Searle, and Penrose were and are not the only doubters, but they do represent the best-known hype alerters. They are all essentially correct that AI cannot just rely on algorithmic, symbolic processing, and that consciousness is not possible purely within the skimpy and restricted confines of a Turing Machine (or a Chinese room . . . ). In the 1950s and 60s, AI folks thought it was possible, but now we know better—about computing, about the brain, the mind, and evolutionary processes. So far, no-one of the stature of a Dreyfus, Searle, or Penrose has risen to challenge the more informed, modern version of strong AI that I support.

The modern view incorporates the tremendous strides made in the past two decades in the computer and cognitive sciences (particularly neural networks, cellular automata/genetic algorithms, evolutionary cellular systems, autonomous agents, and quantum computing); in our knowledge of the brain, reasoning, and emotion and their roles in intelligence, mind, and consciousness; in computing power; in robotics; and in the changes in approach to AI made by the AI community. Whether the underlying operations of these various techniques and concepts are expressible algorithmically (thereby vindicating the strong AI school) remains unresolved scientifically, and Penrose is right to insist on more research. I don’t want to discourage the philosophers, though. They help keep folks like me honest—or at least, careful!

I close with a personal observation. My wife and I became acquainted with an elderly woman afflicted by Alzheimer’s Disease. Let’s call her Mrs. X. Whenever we would meet her, Mrs. X would hone in on my wife and ask: "Where are you from?"

"Flom Japan," says my wife, who never has figured out the L and the R thing. "Oh, how interesting!" came the reply; "I was in Okinawa with my husband in the U.S. Army." Then she would recount specific events and impressions to do with Japan. A few minutes after such a conversation ended, Mrs. X. would again engage my wife in conversation. "Where are you from?" she would ask. "Flom Japan," my wife would reply. "Oh, how interesting! . . ." And the entire conversation would be repeated almost verbatim. Several times in an evening, and on more than one evening.

I don’t know much about Alzheimer’s, and I don’t mean to sound heartless, for this was indeed a tragic situation; but I do know an algorithmic subroutine locked in a loop when I see one.

Until next week,

NEXT WEEK: So What? So what if we are basically just machines? So what if machines do get smarter than us?

Previous Features